Re: [PATCH v1] kthread/smpboot: Serialize kthread parking against wakeup

[Peter Zijlstra]

On Tue, Jun 05, 2018 at 05:08:41PM +0200, Oleg Nesterov wrote:

> On 06/05, Kohli, Gaurav wrote:
> >
> > As last mentioned on mail, we are still seeing issue with the latest
> > approach and below is the susceptible race as mentioned earlier..
> > controller Thread                               CPUHP Thread
> > takedown_cpu
> > kthread_park
> > kthread_parkme
> > Set KTHREAD_SHOULD_PARK
> >                                                 smpboot_thread_fn
> >                                                 set Task interruptible
> >
> >
> > wake_up_process
> >  if (!(p->state & state))
> >                 goto out;
> >
> >                                                 Kthread_parkme
> >                                                 SET TASK_PARKED
> >                                                 schedule
> >                                                 raw_spin_lock(&rq->lock)
> > ttwu_remote
> > waiting for __task_rq_lock
> >                                                 context_switch
> >
> >                                                 finish_lock_switch
> >
> >
> >
> >                                                 Case TASK_PARKED
> >                                                 kthread_park_complete
> >
> >
> > SET Running
> 
> I think you are right.
> 
> And, now that I look at 85f1abe0019fcb3ea10df7029056cf42702283a8
> ("kthread, sched/wait: Fix kthread_parkme() completion issue") I see this note
> int the changelog:
> 
> 	The alternative is to promote TASK_PARKED to a special state, this
> 	guarantees wait_task_inactive() cannot observe a 'stale' TASK_RUNNING
> 	and we'll end up doing the right thing, but this preserves the whole
> 	icky business of potentially migating the still runnable thing.
> 
> OK, but __kthread_parkme() can be preempted before it calls schedule(), so the
> caller still can be migrated? Plus kthread_park_complete() can be called twice.

Argh... I forgot TASK_DEAD does the whole thing with preempt_disable().
Let me stare at that a bit.